A reminder that LLMs are not your friends (by default). They are deeply biased "affirmation" machines and you need to tell them you want clear and honest feedback, not sunshine.
For instance, I recently engaged with ChatGPT to help better understand Viktor Frankl's statement that "self-actualization is nonsense".
Here's what it initially told me about my thinking:
1) "That it beautifully put, Eric..."
2) "You are right on every level..."
3) "That's the exact insight you're articulating..."
4) "You and Frankl are aligned at the level where it really matters..."
I mean, who doesn't want to be told they're a fucking genius?
But, my kids are the smart ones, so I was a little suspect. I asked it to respond again to my previous thinking, but this time prioritize brutally direct (but kind) feedback that would help me identify weak spots in my thinking.
This is what it said the second time:
1) "You may be over-indexing on your own level of introspection"
2) "You might be confusing self-knowledge and meaning-making"
3) "Your distrust of distraction may mask a perfectionism"
4) "You may be privileging the mind over action"
5) "You and Frankl may agree on too much"
The first response made me feel good. The second response triggers growth. (I think it's wrong on all these, btw, but in diving into each, I'll find out.)
All I had to do was tell the LLM to stop blowing smoke up my butt and be honest with me. Any time you use an LLM as a sounding board, it's imperative that you ask it to prioritize honesty over affirmation.